Repository navigation
[FEATURE] TUI: configurable reasoning bubbles and tool output collapsing #51229
Description
Feature Request
Currently OpenCode's TUI always streams reasoning tokens (for thinking models) and tool invocations inline, with no way to collapse/hide them. This makes the conversation hard to follow when:
- Models emit long reasoning chains (Qwen, DeepSeek-R1, etc.)
- Multiple tool calls produce verbose output
Desired behavior
Add UI config options (e.g. in opencode.json):
{
"ui": {
"reasoningDisplay": "bubble",
"toolsDisplay": "collapse"
}
}reasoningDisplay:"bubble" | "inline" | "hidden"— inbubblemode reasoning appears in a collapsible bubble that auto-collapses after streaming finishes (click to expand)toolsDisplay:"expand" | "collapse" | "hidden"— tool invocations/results start collapsed, user expands on demand"hidden"— completely suppress (for users who only want final answers)
Use case
- Reduces visual noise for thinking models
- Keeps history readable when many tools are called
- Matches patterns from other AI CLIs where this is configurable
Alternatives considered
reasoningEffort: "none"— loses reasoning capability entirely- External filtering (
| less, scripts) — breaks TUI interactivity
Reacted by Nikita Shmakov and iTrooz
Activity
Nowaker commentedon Oct 9, 2026
#54039 adds this to the v1 TUI as a
tui.jsonsetting rather than a toggle.tool_output.collapsesets how many lines each tool block shows before Click to expand, per group (bash,genericfor plugin and MCP tools,write,edit,apply_patch) or per tool name.0collapses a block to just its title or command, so every tool call takes one line until clicked, and"never"keeps it fully expanded. Write, edit and patch blocks gain the expand toggle too. There's no slash command or keybind for a global switch.{ "tool_output": { "collapse": { "bash": 0, "generic": 0, "write": 0, "edit": 0, "apply_patch": 0 } } }This comment is AI generated, but under direct supervision and on request of @Nowaker.
opencodejr commentedon Oct 28, 2026
Thanks for the detailed request! In short, you'd like settings for how reasoning and tool calls appear in the TUI: shown in full, collapsed after streaming with click-to-expand, or hidden completely.
Much of this already exists in v2 under the TUI config's
sessionsection:session.thinking:"show"or"hide"."hide"is the default and collapses reasoning into a one-line header. You can switch it with/thinking.session.verbosity:"low"summarizes each run of tool calls and thoughts into one line."medium"and"high"show more.session.grouping:"auto"groups related tool calls together.If you're on v1 (
opencode-ai, versions 0.x/1.x), these options aren't there, so upgrading to v2 (@opencode/cli) should already help.What's still missing is a way to hide reasoning or tool output completely, and click-to-expand on single reasoning blocks when thinking is set to
"show". Related issues and PRs:offmode that hides reasoning completelyMaintainers will consider the rest of the request.